Study On The Methods Of Multi-channel Speech Enhancement

In real life, Speech is inevitable corrupted acoustically by ambient noise which leads to rapid deterioration on the performance of digital voice processor. Therefore, speech enhancement as a measure of pretreatment is an effective way to solve the problem. Over the years, scholars have been making unremitting efforts to find various ways of speech enhancement methods. Methods with single microphone restricted by the number of microphone cannot give us satisfying enhancement results. Methods with microphone array incorporate both spatial and spectral information. Hence, this is better performance than the traditional way, i.e. speech enhancement methods with single microphone.Based on the former research achievements, this thesis makes a research on speech enhancement methods with microphone array as following three aspects:Study on speech enhancement method based on phrase error and post-filter. According to the time difference of arrival of the speech source of interest recorded by the close-by microphones, first, the noisy speech signals are partitioned into segments, by using the phrase error of the two close-by microphones in each segments, we can construct scaling coefficients to preprocess the noisy speech, and then we can get the preprocessed signals. Second, with the estimated power spectrum of preprocessed signals, we can construct the post-filter to further suppress the noise. The noisy speech signals produced by computer are used to test the performance of the proposed method. Simulation results show the effectiveness of the proposed method in diffused noise field and incoherent noise field.Study on speech enhancement method for the reverberation and coherent noise field environments. As for the environments, first, by using blind source separation and spectral subtraction methods to suppress correlated noises in the noisy speech to gain a reverberation speech. Finally, an inverse filter and spectral subtraction are employed to minimize the influence of the reverberation. Simulation results show the effectiveness of the proposed method.Study on some beamformer speech enhancement methods, and a modified speech enhancement method of Generalized Sidelobe Canceller (GSC) is proposed. It could be implemented by estimating the transfer function ratios and processing the input data using Forward-backward filtering. Then a blocking matrix (BM) and the well-known Normalized Least Mean Square (NLMS) algorithm are used to suppress correlated noises further. The noisy speech signals produced by computer and recorded in an actual room acoustic environment are used to test the performance of the proposed method. Simulation results show the effectiveness of the proposed method in coherent noise environments.